MCP Server - priorwork.fyi

🔌 What is this?

priorwork-mcp is a small Model Context Protocol server that puts semantic search over 128,700+ papers from major ML, NLP, and Computer Vision conferences (NeurIPS, CVPR, ICML, ICLR, ACL, EMNLP, ICCV, AISTATS, and more) directly into your agent's tool list — no browser tab, no copy-pasting abstracts. Ask your agent to find prior work on an idea and it calls the tool itself.

⚡ Quick Start

One command, no API key required to start — it defaults to a shared public key.

# Claude Code
claude mcp add priorwork -- uvx priorwork-mcp

For Cursor or any other MCP client, add this to your MCP configuration:

{
  "mcpServers": {
    "priorwork": {
      "command": "uvx",
      "args": ["priorwork-mcp"]
    }
  }
}

Runs via uvx, so the only prerequisite is uv — nothing to install or keep updated yourself.

🧰 Tools

Tool	Description
`search_papers`	Semantic search from a text query (a short descriptive sentence or two works slightly better than a few keywords) — returns ranked papers with authors, venue, year, similarity score, download URL, and full abstract. Pass `summarize=true` to add a structured per-paper LLM summary (problem, core insight, method, key results, positioning).
`search_by_pdf_url`	Give it a public PDF link (e.g. an arXiv PDF) and get semantically similar papers from the index.
`search_research_topics`	A different corpus: open research problems mined from the limitations authors state in their own papers — gaps several independent papers ran into that nobody has closed. Describe what you want to work on as prose; keywords match nothing here, since each problem is stored as a synthetic abstract. Returns the problem, why it matters, what changes if it works, the risk, and concrete approaches.
`list_indexes`	List the available paper indexes and their sizes.

The topics artifact is cached locally and refetched only when the server's artifact id changes, so repeated searches cost one request rather than a 4.6 MB download each.

⚙️ Configuration

Everything is optional — configured via environment variables, passed with -e KEY=value in claude mcp add, or an "env" block in JSON configs.

Variable	Required	Description
`PRIORWORK_API_KEY`	optional	Your own key from below. Defaults to the shared public free-tier key, rate-limited across every default install.
`PRIORWORK_API_URL`	optional	Override the API base URL. Defaults to `https://priorwork.fyi`.

Public Free Key (default)

pw_live_65d65b6a1f42920fda777e306f0005c0b9141bba4327dff203be010c1f34be72

The shared key is capped at 500 requests/day in total across everyone using the default install, so if you're hitting limits, set PRIORWORK_API_KEY to a key of your own — see the API docs for the same key, unlimited for your own use.

# Claude Code, with your own key
claude mcp add priorwork -e PRIORWORK_API_KEY=your_key_here -- uvx priorwork-mcp

💬 Example

"Find prior work on spatiotemporal forecasting with graph neural networks"

1. Taming Local Effects in Graph-based Spatiotemporal Forecasting — Andrea Cini, et al. (2023 · NeurIPS) [score: 0.802] 2. FourierGNN: Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective — ...

This is exactly the check worth running before starting new work, or before a submission — a topic can rack up over a thousand papers in a single year (see the paper timelapse), more than any one person can track by hand.